<p>The fewer samples in tail classes in comparison to head classes make them harder to recognize, leading to inferior classification performance for tail classes in long-tailed image classification task. By considering that head class samples contain abundant feature information and there is a similarity among samples of the head and tail classes, we propose a Multi-Samples Features Transfer and Preserving (MSFTP) module to fully use the head class samples. The MSFTP module consists of a Multi-Samples Features Transfer module (transfer module) and a Multi-Samples Features Preserving module (preserving module). The transfer module transfers feature information of samples from the head classes to the tail classes, shifting the model’s focus. The preserving module tries to preserve exclusive feature information for head class samples, aiming to maintain their classification accuracy during the feature transfer. The two modules work together to promote the classification performance of tail classes. Our proposed method can be easily incorporated with existing frameworks to further enhance classification performance. We carry out a significant number of experiments across the CIFAR10-LT, CIFAR100-LT, SVHN-LT, and CINIC10-LT datasets and achieve promising result.</p>